India’s digital payments infrastructure is beginning to move beyond AI assistants and analytics toward systems that can actually coordinate complex operational work. The National Payments Corporation of India (NPCI) on September 9 launched two agentic AI platforms, AiNxt and Agentic Orchestration & Messaging (AtOM), at the Bharat AI Zone of Global Fintech Fest 2026 in Mumbai.
The two platforms target different layers of the problem. AiNxt is designed as an enterprise platform for building and deploying AI agents, while AtOM is focused on automating the integration, change-management, partner-onboarding and certification workflows that sit behind payment ecosystems such as UPI.
NPCI Is Putting Agents Behind the Payment Infrastructure
The distinction matters. NPCI is not announcing an AI chatbot that replaces a UPI app or asking consumers to hand over their payments to autonomous software. Instead, the organisation is applying agentic AI to the systems and workflows that banks, payment service providers and technology partners use to connect with and operate around UPI.
That is a potentially important shift because UPI is already operating at enormous scale. Reuters reported that UPI processed 24.51 billion transactions worth roughly $314 billion in August 2026. The challenge at that scale is not simply processing another transaction; it is managing the growing number of banks, apps, service providers, integrations, certifications, technical changes and compliance requirements surrounding the network.
What Is AiNxt?
AiNxt is the broader of NPCI’s two new platforms. It is positioned as an enterprise-grade environment where organisations and developers can create, test and deploy customised AI solutions and agents.
One of its more notable features is Bring Your Own Models (BYOM). Rather than forcing organisations to use one underlying AI model, AiNxt is designed to let users work with models of their choice and build customised agentic applications around them. NPCI is also positioning the platform as open source, giving organisations the ability to access, modify and extend the software.
The AiNxt suite includes four components: AiNxt OS, AiNxt Code, AiNxt CLI and AiNxt Enterprise. Together, they are intended to cover different parts of the development workflow, from no-code and low-code experimentation to more developer-centric work inside integrated development environments and command-line tools.
For financial institutions, that model flexibility can be useful. Banks and payment companies often have to balance capability with regulatory requirements, data controls and existing infrastructure. A platform that supports multiple models can potentially give engineering teams more freedom to choose where inference happens and how an AI system is integrated into internal workflows.
AtOM Targets the Harder Part of UPI: Coordination
AtOM — short for Agentic Orchestration & Messaging — is more tightly connected to UPI operations.
NPCI says AtOM can coordinate complex processes across design, development, integration, partner onboarding and certification through a single auditable workflow. In practice, that means a bank or payment service provider could use the system to automate parts of the process required to adopt a new UPI capability rather than coordinating every stage manually across multiple teams and organisations.
The platform is built around JSON-RPC 2.0, a lightweight protocol for remote procedure calls. NPCI says AtOM’s architecture is designed to support machine-readable and digitally signed interactions, creating a more structured record of actions and helping with compliance and audit readiness.
That could become increasingly valuable as the UPI ecosystem adds more capabilities. Every new feature has to work across banks, apps, payment processors and other ecosystem participants. The technical work is only one part of the challenge; testing, certification, readiness tracking and coordination can become bottlenecks as the number of participants grows.
Why This Is Different From a Typical AI Assistant
Agentic AI is often described simply as AI that can take actions instead of only answering questions. The more meaningful difference is the workflow around those actions.
A conventional AI assistant might explain how a UPI integration works. An agentic system can be designed to take a sequence of steps: inspect requirements, call software tools, move information between systems, check the result, and continue toward a defined objective. AtOM is an example of that concept being applied to enterprise infrastructure rather than consumer chat.
That does not mean NPCI is handing an AI unrestricted control over UPI. The platform’s emphasis on auditable workflows and digitally signed interactions points in the opposite direction: automation is being placed inside a structured process where actions can be recorded, checked and governed.
What It Could Mean for Banks and UPI Apps
The immediate opportunity is reducing the operational friction around UPI changes.
Today, introducing a new payment feature can involve product teams, developers, security teams, compliance staff, testing teams and external partners. Each organisation may have its own systems and processes. A shared orchestration layer could reduce some of the repetitive coordination between those groups.
For banks and payment service providers, the potential benefit is faster deployment of new UPI capabilities without proportionally increasing the amount of manual integration work. For NPCI, the benefit could be greater consistency in how ecosystem participants interact with the infrastructure.
There is also a governance angle. The more financial infrastructure depends on automated software agents, the more important it becomes to know what an agent did, which systems it accessed and whether the result complied with predefined rules. NPCI’s emphasis on auditable and digitally signed interactions suggests that traceability is being considered alongside automation rather than added later.
India’s Agentic AI Push Is Moving Into Digital Public Infrastructure
NPCI’s move is significant because UPI is not an ordinary commercial app. It is part of India’s digital public infrastructure and connects a huge ecosystem of banks, payment apps and merchants.
NPCI has already been building a broader AI strategy for payments. Earlier in 2026, the organisation announced a collaboration with Nvidia to advance sovereign AI capabilities for India’s payments ecosystem, including work around open-weight models and payment-specific AI infrastructure. The new AiNxt and AtOM platforms show another side of that strategy: turning AI capability into operational software that can sit around the payment rails.
The approach also fits the wider agenda at Global Fintech Fest 2026, whose programme places agentic AI, tokenisation and quantum technology among its central themes.
The Bigger Question: How Much Autonomy Should Payments Infrastructure Have?
The most interesting part of NPCI’s announcement may not be what AiNxt and AtOM can automate today, but where the architecture could lead.
As AI agents become better at using APIs and coordinating software tools, payment infrastructure is an obvious environment for both experimentation and caution. Agents could eventually help with reconciliation, fraud investigation, merchant operations, customer support, compliance workflows and technical change management.
But payments also demand unusually strong controls. A software agent making a mistake in a document workflow is inconvenient; an agent making an incorrect financial action can create a direct monetary loss. That makes authentication, permissions, audit trails, human oversight and clear boundaries especially important.
For now, NPCI is starting on the infrastructure side rather than giving autonomous agents direct control over consumers’ money. That is a more conservative starting point, but potentially a more consequential one. If AtOM can make UPI integrations faster and more reliable while keeping those processes auditable, agentic AI could become part of the plumbing of India’s payment ecosystem long before consumers notice an AI agent making a payment on their behalf.
What Comes Next
AiNxt and AtOM represent an early attempt to turn agentic AI from a productivity concept into infrastructure for a system that processes billions of transactions each month. Their success will depend less on how impressive the AI sounds and more on whether banks and payment partners can use the platforms to reduce integration time, improve reliability and maintain strong governance.
If that happens, NPCI’s experiment could become a useful model for other forms of digital public infrastructure: use AI to automate the complicated work behind the service while keeping the underlying system observable, interoperable and accountable.
